Comparação de métodos no ajustamento de curvas de lactação de bovinos por meio de Simulação

Abstract

Lactation curves represent milk production of a dairy dam as a function of time. Since such curves vary randomly among individuals, due to environment and genetic factors, the mixed model known as random regression is suitable for fitting such curves, and it was evaluated in this study through data simulation. Milk production of five animals in five ages was simulated in one thousand random and independent data sets, considering three precision levels, variation due to the model in relation to the residual variation of 0,9 (high variance), 0,5 (medium variance) and 0,1 (low variance) and three average degrees of relationship. Random regression was used in two ways, admitting or not that the variances of the model (genetic and of environment) were known, and compared to the ordinary least squares method. Results showed that, under low variance, the models had similar fits, with regard mean square error and mean absolute deviation. The level of precision was more influential than the degree of relationship on relative performance of models. In general, random regression with known variances yielded the most precise predictions, followed by random regression estimating such variances. Results suggest that random regression is potentially useful under high environmental heterogeneity, even if variances are unknown.